On Thu, Mar 3, 2011 at 3:54 AM, Chris Caudle <ch...@chriscaudle.org> wrote:
> Has anyone found reference which derives the worst case clock jumps to > expect when using PTP (IEEE1588-2008) and how to derive a phase noise > spectrum from that? > As Magnus says, this is highly dependent on implementation. For "standard" implementations, in which a non-controllable oscillator is used in the PTP clocks, the digital servo has a sampling rate which varies with network latencies (which themselves vary with network load), so it is virtually impossible to give a precise answer to your question. > I'm looking at a protocol which uses 1588-2008 to create a common clock > between networked devices, and those devices are supposed to derive other > clocks (e.g. sampling clocks) synchronized to the 1588 clock. We have exactly that requirement at CERN, and the White Rabbit project ( http://www.ohwr.org/projects/white-rabbit/wiki) is our answer to it. The main idea is very simple: PTP does not tell you anything about the nature of the oscillator you should use in a PTP device. Instead of using a non-controllable oscillator, we derive our clock signal from the incoming Gigabit Ethernet data stream. The rx_clock from the PHY is always good, even in the absence of Ethernet packet transfer. By doing that, we have a system which is easier to analyze (constant sampling rate) and we require much less PTP traffic since PTP now only has to deal with very slow thermal effects (like fiber length changes). To our knowledge, and as per the ISPCS 2010 plug fest, White Rabbit is currently the most precise PTP implementation. We are preparing a document with a phase noise PSD plot for reference and recovered clocks using a 5-km fiber link. Typical clock signal jitters are below 100 ps and UTC-transfer accuracies are below 1 ns. For a quick overview of WR you can see http://svn.ohwr.org/white-rabbit/trunk/documentation/presentations/WR_Tomek_BE_TechCommittee/javier_ni_talk.pdf Cheers, Javier _______________________________________________ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.